Influence of Arctic sea-ice variability on Pacifictrade windsCharles F. Kennela,b,1 and Elena Yulaevac

aScripps Institution of Oceanography, University of California San Diego, La Jolla, CA 92093; bCentre for Science and Policy, University of Cambridge, CB23BU Cambridge, United Kingdom; and cSan Diego Supercomputer Center, University of California San Diego, La Jolla, CA 92093

Contributed by Charles F. Kennel, December 3, 2019 (sent for review October 25, 2017; reviewed by Ivana Cvijanovic, Jennifer Francis, and James E. Overland)

A conceptual model connecting seasonal loss of Arctic sea ice tomidlatitude extreme weather events is applied to the 21st-centuryintensification of Central Pacific trade winds, emergence of CentralPacific El Nino events, and weakening of the North Pacific AleutianLow Circulation. According to the model, Arctic Ocean warmingfollowing the summer sea-ice melt drives vertical convection thatperturbs the upper troposphere. Static stability calculations showthat upward convection occurs in annual 40- to 45-d episodes overthe seasonally ice-free areas of the Beaufort-to-Kara Sea arc. Theepisodes generate planetary waves and higher-frequency wavetrains that transport momentum and heat southward in the uppertroposphere. Regression of upper tropospheric circulation data onSeptember sea-ice area indicates that convection episodes producewave-mediated teleconnections between the maximum ice-lossregion north of the Siberian Arctic coast and the IntertropicalConvergence Zone (ITCZ). These teleconnections generate oppo-sitely directed trade-wind anomalies in the Central and EasternPacific during boreal winter. The interaction of upper tropospherewaves with the ITCZ air–sea column may also trigger Central Pa-cific El Nino events. Finally, waves reflected northward from theITCZ air column and/or generated by triggered El Nino events maybe responsible for the late winter weakening of the Aleutian LowCirculation in recent years.

Arctic sea ice | decadal variability | Pacific trade winds | Central Pacific ElNino | Aleutian Low

Forty years ago, Manabe and Stouffer (1) showed that theArctic surface temperature increases significantly faster than

the global mean surface temperature (GMST) in response to in-creasing greenhouse-gas concentrations. Two feedback mecha-nisms, from greenhouse warming due to water vapor transportedinto the Arctic and from increased absorption of sunlight due toseasonal loss of Arctic sea ice and snow cover, are responsible forthis “Arctic amplification.” The balance of the two mechanismschanged near the end of the 20th century. The change in Arcticclimate that started at that time proved due primarily to increasein summer losses of sea ice (2). That change was subsequentlyrelated to changes in the duration and pattern of midlatitude ex-treme weather events (3, 4). Here, we investigate whether theincreased loss of summer Arctic sea ice was also related to thechanges in trade-wind strength and El Nino behavior observed inthe Tropical Pacific.

Recent Change in Arctic and Tropical Pacific ClimatologyTwo assessments of Arctic climate published 7 y apart illustratedhow rapidly Arctic climate dynamics had shifted. The 2004 ArcticClimate Impact Assessment (ACIA) (5) created a 20th-centurybaseline to which the 21st-century climate could be compared.ACIA’s 2011 sequel, the Snow, Water, Ice, and Permafrost As-sessment (SWIPA) (6), found that a structural transformationhad taken place in the 7 y separating the two assessments. Therate of increase of Arctic mean surface temperature between2004 and 2011 was the largest on record. Warming had beengreater over land before 2004, but between 2004 and 2011, the

larger warming rate was over ocean. This was an indication thatthe sea-ice component of Arctic amplification had strengthened.Satellite measurements of the earth’s radiation budget con-

firmed that the Arctic’s absorption of short-wave solar radiationincreased during the period documented by ACIA (7). The sum-mer sea-ice melt uncovers areas of dark ice-free ocean that absorb93% of the short-wave solar radiation reaching the surface,whereas white ice reflects 50 to 80% back to space. Snowmelt hasa comparable effect (8). The Arctic surface heating by solar short-wave radiation increased by 2 W/m2 between 1979 and 2000; be-tween 2001 and 2011, solar forcing increased by another 4 W/m2,consistent with the larger decreases in sea-ice area after the turn ofthe 20th century. When spread over the surface of the earth, theincrease in Arctic forcing from 1979 to 2011 was about one-fourthof the increase due to anthropogenic carbon dioxide during thesame period. This and other lines of evidence indicated that thelarge summer losses of sea ice were contributing to changes inArctic climate (9) judged so fundamental that a 2013 review article(10) was entitled “The Arctic Shifts to a New Normal.”The Arctic and the globe had embarked on diverging surface-

temperature trajectories at about the same time. In colloquialterms, global warming decelerated, while Arctic warming accelera-ted. The period of decelerated GMST increase, which has beencalled the “hiatus” in global warming (11), provoked conflictedpublic (12) and policy (13) debates. Scientifically, by 2015, it becameclear that the 5th Coupled Modeling Intercomparison Project

Significance

By 20th-century standards, the Central Pacific trade winds thatdrive the El Nino–Southern Oscillation feedback system to in-stability have been unusually strong in the 21st century. Theannual summer melts of Arctic sea ice are up to twice as largein area as in the 20th century. Arctic sea ice, upper atmosphericcirculation, surface wind, and sea-surface temperature dataprovide evidence that upper troposphere transport processesconnect the increased summer losses of Arctic sea ice to thetrade-wind and Central Pacific El Nino events characteristic ofthe present climate state. These results add to the evidencethat loss of Arctic sea ice is having a major impact on climaticvariability around the world.

Author contributions: C.F.K. and E.Y. designed research; C.F.K. and E.Y. performed re-search; C.F.K. and E.Y. analyzed data; and C.F.K. wrote the paper.

Reviewers: I.C., Lawrence Livermore National Laboratory; J.F., Rutgers University; andJ.E.O., National Oceanic and Atmospheric Administration Pacific Marine EnvironmentalLaboratory.

The authors declare no competing interest.

This open access article is distributed under Creative Commons Attribution-NonCommercial-NoDerivatives License 4.0 (CC BY-NC-ND).

Data deposition: Derived data that support the findings of this paper and associatedcodes are available in the GitHub repository at https://github.com/eyulaeva/PNAS-2017-17707.1To whom correspondence may be addressed. Email: ckennel@ucsd.edu.

This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1717707117/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1717707117 PNAS Latest Articles | 1 of 11

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Apr

il 16

, 202

0

(CMIP5) climate model ensemble did not account for the behaviorof the GMST, tropical trade winds, and Arctic sea ice during thehiatus. However, subsurface ocean heat content (OHC) did in-crease at the rate expected from the increase in greenhouse-gasconcentrations (14). Anthropogenic forcing had evidently not di-minished, but its effects were distributed differently within the cli-mate system. Empirical studies of decadal climate variabilityprovided a candidate explanation. Pacific Ocean sea-surface coolingfollowing the 1998 to 2000 switch in phase of the InterdecadalPacific Oscillation (IPO) may have offset the previous rate of in-crease of GMST (15) in a way that the CMIP5 ensemble did notcapture (though particular models were more successful).The El Nino event of 1997 to 1998, the largest in the instru-

mental record, marked an abrupt change in Tropical Pacific climate.In the year following that event, unusually strong trade winds in theCentral Pacific (CP) (16), a long period of La Nina-like cooling ofthe CP and Eastern Pacific (EP) sea surface (17, 18), and increasedocean heat sequestration in the CP and Western Pacific (19) allbegan. A 2nd type of El Nino event, the CP El Nino (20–25),found only once in instrumental data before 1990 (26), became thedominant type of El Nino event. Proxy data show that the ratio ofCP to EP events was the largest in four centuries (27). CP El Ninoevents produced long-distance teleconnections (23) that differedfrom those of the EP El Nino, and an Indian Ocean counterpart tothe CP El Nino—the Ningaloo Nino—was identified (28).The first decade of the 21st century, 2000 to 2010, was called a

“decade of weather extremes” (29, 30). The severe North Amer-ican winter of 2009 to 2010 was traced to Arctic air invading themidlatitudes in deep meanders of the jet stream (31). The Moscowheat wave and Pakistani flooding in the summer of the same yearwere also attributed to a slowly moving meander (32). By 2012,researchers were debating whether the changing pattern of mid-latitude extreme weather events and the increased summer lossesof Arctic sea ice (3) were related. Though the scientific debateremains vigorous (33–35), the case is strengthening that the largermelts of Arctic sea ice are altering the spatial pattern and intensityof midlatitude extreme weather events (36, 37).All in all, a change in decadal climatic behavior occurred

during the last 2 to 3 y of the 20th century. It required the firstdecade of the 21st century to confirm the new trends, but thereduced rate of GMST increase, increased summer losses ofArctic sea ice, strengthened CP trade winds, increased frequencyof CP El Nino events, cooled sea surface in the equatorial CP andEP, and changed midlatitude extreme weather patterns proved tobe components of the new climate configuration. Many of the newfeatures are consistent with a positive-to-negative switch in phaseof the IPO, but the magnitude of Arctic sea-ice loss and thestrength of the CP trade winds are unprecedented in the in-strumental record and unique to the present IPO phase.

Conceptual Model, Methods, and DataClimate-model ensembles are an objective way to assess themultidecadal impacts of anthropogenic greenhouse-gas emis-sions on global climate. However, the CMIP 5 ensemble did notaccount for the reduction in GMST growth rate (38), trade-windintensification (16), and increased seasonal losses of sea ice afterthe 1997 to 1998 El Nino (39, 40). Until these shortcomings arediagnosed, questions will remain about how models portray cli-mate variability on the decadal time scale. Using observationaldata to trace connections between physical processes providesindependent information, but it has a cost. Causal ambiguitiesmust be resolved by assuming the time order of those processes.This is done here by postulating an extension of the conceptualmodel proposed to explain the influence of sea-ice variability onmidlatitude extreme weather:

1) Vertical convection is generated in episodes of about 1-moduration over the ice-free areas in the Pacific-facing sector of

the Arctic Ocean in early boreal fall (Initiation, Duration, andSpatial Extent of Convection Episodes);

2) Vertical convection episodes stimulate a planetary wave anda southward-propagating higher-frequency wave train in theupper troposphere that interact with the Indo-Pacific Inter-tropical Convergence Zone (ITCZ) in December (P2 ArcticITCZ-Teleconnections, Difference between P1 and P2 Telecon-nections, and Southward Advance of P2 Teleconnections);

3) The interaction of the wave and wave train with the ITCZ aircolumn creates conditions favorable to triggering CP El Ninoevents (Westerly Wind Bursts and the CP El Nino);

4) A reflected northward propagating wave and wave train iscreated in the interaction of the Arctic-origin waves withthe ITCZ air column (Westerly Wind Bursts and the CP ElNino and Arctic Sea Ice and the Aleutian Low Circulation);

5) The Aleutian Low Circulation responds to the passage of thereflected waves over the North Pacific in February (Arctic SeaIce and the Aleutian Low Circulation);

6) The cumulative impact of repeated large annual losses ofSeptember Sea ice shapes the climatology of the NorthernPacific after 1999 (Recent Change in Arctic and TropicalPacific Climatology).

The sequence of events 1 to 5 repeats annually, but the in-tensity and spatial structure of the annual episodes of convectionvary due to contingent factors like clouds, storms, and watervapor content in the Arctic atmosphere (41, 42). Snow cover,surface roughness, and melt-water ponding affect sea-ice albedoand thus its melt rate, influencing when and where convectionepisodes start and stop. The amplitude of a convection episodemay be thought of as a product of intensity, area, and duration;despite the seasonal regularity of solar insolation, the timing andamplitude of Arctic convection episodes vary from year to year.This paper investigates how tropical Pacific climate adjusts to

changes in amplitude of Arctic convection episodes. Compo-nents of the climate system that respond to convection episodesmay be identified by regression on September sea-ice area(SSIA). The late 20th century was a “high ice” period (smallersummer sea-ice loss) and the 21st century a “low-ice” period(larger loss); the data from the 2 periods are binned separately tohighlight the change in relationship between Arctic sea ice andtropical climate noted in Recent Change in Arctic and TropicalPacific Climatology. Many models of the impacts of Arctic sea-icevariability on extra-Arctic climate have not been energy con-serving. By comparing the results of “low-ice” and “high-ice”energy-conserving climate models, Cvijanovic et al. (43) foundthat the Aleutian Low Circulation in the North Pacific could beinfluenced by Arctic Sea-ice loss via a 2-step teleconnection:Arctic to Tropical Pacific and Tropical Pacific to North Pacific.These results add to the case that the recent changes in Arcticsea ice are having important effects on midlatitude climate.For this paper, however, the 2-step teleconnection is the im-

portant idea; Matsumura and Kosaka (44) propose a related pic-ture. The view emerging from Cvijanovic et al. is consistent withsteps 1, 2, 4, and 5 in the sequence above. Their results lend con-fidence to the use here of that sequence of steps to interpret sta-tistical analyses that viewed singly are ambiguous. Viewed together,the empirical analyses presented here support their picture, identifyclimatic processes that respond to SSIA variability, and charac-terize the differences between the 2 high-ice decades before theturn of the 20th century and the 2 low-ice decades that followed.The properties and limitations of the public-domain data used

here have been extensively documented. Atmospheric variablesand sea-surface temperature (SST) were derived from theNational Centers for Environmental Prediction (NCEP)/NationalCenter for Atmospheric Research (NCAR) reanalysis monthlymeans (45, 46) or from the daily European Centre for Medium-Range Weather Forecasts interim Reanalysis (ERA-Interim) (47).

2 of 11 | www.pnas.org/cgi/doi/10.1073/pnas.1717707117 Kennel and Yulaeva

Dow

nloa

ded

by g

uest

on

Apr

il 16

, 202

0

The ERA-Interim product has proven superior for high-latitudestudies (48, 49). Arctic Sea Ice area was calculated by using theNational Oceanic and Atmospheric Administration (NOAA)/National Snow and Ice Data Center (NSIDC) Climate DataRecord of Passive Microwave Sea Ice Concentration, Version3 (50, 51). Sea-surface height (SSH) relative to the geoid wasobtained from NCEP Global Ocean Data Assimilation System(GODAS) Reanalysis data (52) provided by the NOAA/OAR(Oceanic and Atmospheric Research) Earth System ResearchLaboratory’s Physical Sciences Division, Boulder, CO, fromtheir website at https://www.esrl.noaa.gov/psd/. Derived datathat support the findings of this paper and associated codesare available in the GitHub repository at https://github.com/eyulaeva/PNAS-2017-17707.

The 1997 to 1998 Transition in Equatorial Pacific ClimateThe Hovmuller diagrams (53) in Fig. 1 display the longitude andtime dependences of the monthly anomalies relative to 1980 to1998 climatology of SSH, SST, sea-level pressure (SLP), and10-m zonal wind velocity (SSU) averaged over ±7.5° latitude. SSTsexceeding 26 °C to 28 °C are required to drive the deep atmo-spheric convection generated by storms over tropical oceans (54,55). The superposed black lines convey the time-dependence ofthe longitude of the 29 °C isotherm, which locates the longitudewhere the measured SST averaged over a ±7.5° latitude bandequaled 29 °C. A sharp gradient in sea-surface salinity (SSS) (the“barrier”) has been found near the 29 °C isotherm and to moveeast and west with it (56, 57). The salinity barrier effectively de-fines the eastern boundary of the Western Pacific Warm Pool(WPWP), and its association with the 29 °C isotherm provides away to track its motions by using SST data, whose time series ismuch longer than that of SSS.The SSH, SLP, SSU, and SST time series changed character

after the 1997 to 1998 El Nino event, which is identified by thepositive SST anomaly in Fig. 1, Lower Left. The trade winds(SSU) strengthened in the year after the event. Strong easterlyanomalies (blue) appeared in the CP (165°W to 150°E) in 1999and in the 2 following years; they were again strong between2007 and 2014 and after the 2015 El Nino event. They were thestrongest in a century-long data record at these times (16). SSHwas generally lower and SLP generally higher in the EP after the1997 to 1998 event. Transient El Nino events characterize thechange in SST behavior in the CP, but more persistent change isclear west of the 29 °C isotherm: SST in the WPWP increasedafter the 1997 to 1998 El Nino event and remained elevated.Easterly trade-wind anomalies in the CP were accompanied by

westerly anomalies in the EP (90°W to 150°W) after 1999. Wecall the combination of opposite-sign anomalies the Trade-WindDipole Anomaly. The 29 °C isotherm is the western boundary ofthe trade-wind dipole anomaly. Inspection of Fig. 1 reveals thatwhen the 29 °C isotherm was in the Western (Central) Pacific,the dipole anomaly was strong (weak), and westerly winds in theWPWP were weak (strong).We define a trade-wind dipole anomaly index by averaging the

NCAR/NCEP 10-m zonal wind velocity over areas N4 and N3,spanning the equatorial CP and EP, respectively, and taking thedifference. N4 and N3 have the same longitudinal extents as theNino4 (150°W to 160°E) and Nino3 (90°W to 150°W) regions(58) used in El Nino–Southern Oscillation (ENSO) forecasting,but their latitudinal extents are increased from ±5° to ±7.5° toinclude the Northern Hemisphere latitude where the ITCZ istypically found in boreal winter. The difference of the N3 and N4averaged 10-m wind velocities, U3 −U4, is the trade-wind dipoleindex.* The Trade-Wind Dipole Anomaly and the Southern

Oscillation Indices are shown to be comparable indicators of ElNino event likelihood in SI Appendix, section S2.To characterize the consequences of the transition in decadal

climate documented here, data will be binned into sets of ap-proximately equal duration preceding and following the transi-tion. The SSIA time series does not provide an unambiguouschoice of transition year; however, the 1997 to 1998 El Ninoevent did mark a definite shift in Equatorial Pacific climaticbehavior. We chose 1998 to be the transition year and defined P1to be the period from 1980 and 1998 and P2 to be the periodfrom 1999 to 2015. P2 to P1 epoch difference maps of the fourvariables in Fig. 1 are presented in SI Appendix, section S1.

P1 and P2 El Nino EventsEastward displacement of the 29 °C isotherm to the CP sets thestage for EP and CP El Nino events (59–61), named according tolongitude of origination. The signature of the EP El Nino is apositive SST anomaly that originates in the EP and spreadswestward, whereas that of the CP El Nino is a positive SSTanomaly that originates in the CP (23) and spreads eastward.The positive SST anomalies (red) in Fig. 1, Lower Left documentEl Nino events that occurred during P1 and P2. There were threeEP (1982, 1987, and 1998) and three CP (1992 to 1995) events inP1 and eight CP events of varying strength, one incomplete EPevent (2014), and one completed EP event (2015) in P2.Both kinds of El Nino event in Fig. 1 occurred at times of

maximum eastward displacement of the 29 °C isotherm. Themaximum displacement longitude was for all practical purposesthe starting longitude for CP El Nino events; the 29 °C isothermretreated westward after the CP events started. By contrast,maximum eastward displacement occurred before one and dur-ing the three other EP events between 1979 and 2017. In all fourEP cases, the 29 °C isotherm returned westward after the events.

Correlation of September Sea-Ice Concentration withDecember Trade-Wind Dipole Anomaly IndexAccording to the conceptual model in Conceptual Model, Meth-ods, and Data, upper tropospheric waves launched during con-vection episodes take 1 to 2 mo to propagate from the Arctic andinteract with the trade winds in the vicinity of the ITCZ. If so,one might expect to find delayed correlations between sea-icearea in September and trade winds in December. Let P1 Sep-tember and P2 September denote the sets of monthly averagedata for all of the months of September in the P1 and P2 timeintervals, respectively, and adopt analogous terminology forother months. Fig. 2, Upper (Fig. 2, Lower) maps the correlationbetween P1 (P2) September Sea-Ice Concentration (SSIC) dataand P1 (P2) December trade-wind dipole index data. Negativecorrelation (blue) implies that an SSIC decrease at that locationin September contributes to an increase in trade-wind dipoleindex in December, and vice versa for positive correlation (red).The sea-ice edge in the Pacific-facing sector of the Arctic (90°W to

180° to 90°E) retreats northward from the coastlines during thesummer ice melt and advances southward toward the coastlinesduring the fall refreeze. In P2 data, the strongest correlations areconcentrated in the region of retreat and advance: a thin blue bandof high negative correlation (r = −0.8) north of the coastline from theBeaufort Sea to the Laptev Sea. SSIC was less than 15% in the re-gion devoid of data between the high correlation band and the coasts.This is a region of nearly complete ice melt (62), where solar heatingof the ocean is intense and upward atmospheric convection is likely.The band of large negative correlation poleward of the no-data re-gion is the ice edge. Since the correlation is negative around theBeaufort–Laptev sector, poleward retreat (equatorward advance) ofthe ice edge would increase (decrease) the trade-wind dipole index.The spatial pattern of September–December correlations was

different in P1 and P2 data. In P1 data, there were no zero-dataregions (that is, sea ice rarely left the coasts) and no regions

*By convention, U positive (negative) corresponds to westerly (easterly) wind direction.Positive (negative) U3 to U4 corresponds to La Nina (El Nino) conditions.

Kennel and Yulaeva PNAS Latest Articles | 3 of 11

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Apr

il 16

, 202

0

where the correlations were as large as in P2 data. In P1 data, theregions bounding the North American and Siberian coasts hadmixed positive and negative correlation, positive (r = +0.6) nearthe Beaufort Sea coast and mostly negative (r = −0.4) near theEast Siberian and Laptev Sea coasts. If ice-edge retreat or ad-vance in a given year were uniform around the Beaufort–Laptevsector, one would expect the change in the P1 trade-wind dipoleindex to be small, since the contributions from the positive andnegative correlation regions would offset one another.In summary, monthly December trade-wind dipole indices corre-

late with SSIC spatial variability in both P1 and P2 data. The P2correlation is particularly strong (r = −0.8) in an annular bandextending from the Beaufort Sea to the Laptev Sea that can beidentified with the sea-ice edge. The strength and location of thehigh correlation band suggests that ice-edge movement to the north(south) in September will correlate with increase (decrease) in trade-wind dipole index in December. The following sections assess thelinks in the chain of processes that connect Arctic sea ice to tradewinds. That chain starts in the region of warmed ocean, where sea iceretreats and advances and vertical convection is initiated in early fall.

Initiation, Duration, and Spatial Extent of ConvectionEpisodesFig. 3 displays maps of the static stability (63) of the atmosphere80 m above the Arctic region selected from a sequence of 11-daverages computed from ERA-interim data center displaced by 5d from August 1 to December 1. ERA-5 data were used to mapconvectively available potential energy in the same format (notshown). The two sets of results both identify a region of in-stability with the seasonally ice-free area of open ocean in thePacific-facing sector (90°W to 180° to 90°E). The 1949 to 2017time history of Pacific Sector open-water area is presented in SIAppendix, section S3.The timings and spatial extents of the unstable regions in P1

and P2 data differed. In both periods, the atmosphere at 80-maltitude was stable (red) in August, unstable (blue) in parts ofSeptember and October, and stable in November and December.The left- and right-hand columns of Fig. 3 Upper Left, Lower Left,Upper Right, and Lower Right show the static stability maps wheninstability first (Left) and last (Right) appeared in P1 (Upper) andP2 (Lower). Instability appeared first and disappeared last nearthe New Siberian Islands, straddling the Laptev and East Siberian

Seas in both P1 and P2 data. Instability started 10 d earlier in P1than in P2 and ended 15 d later in P2. P1’s unstable period lasted40 d, and P2’s 45 d. The regions of instability were located north ofthe North American and Siberian coastlines but differed inpoleward extent in P1 and P2 data. Fig. 3, Upper Center and LowerCenter show the 11-d intervals when the unstable regions achievedthe largest area. The solid lines denote the P1 or P2 average SSIAsea-ice edge at the time of maximum sea-ice loss. The unstableregion extended from the coasts to the sea-ice edge in both P1 andP2 data.Additional assumptions and calculations would be necessary

to convert Pacific-facing sector ice-free area to convection epi-sode amplitude. There is, however, a practical approach totesting hypothesis 2 of Conceptual Model, Methods, and Data:regress on SSIA, which is routinely measured. SSIA is not aperfect proxy for episode amplitude, because contingent factorslike cloudiness (64) and storms (65) affect convection thresholdsand rates. Nonetheless, regression on SSIA does identify extra-Arctic climatic processes that respond to changes in the annualminima of Arctic sea-ice area.†

P2 Arctic-ITCZ TeleconnectionsTo proceed further, it is helpful to describe hypothesis 2 ofConceptual Model, Methods, and Data with more specificity.Convection episodes inject heat and moisture into the uppertroposphere near the end of every sea-ice melt season. The up-ward baroclinic flow is neither spatially uniform nor steady andcreates variability of the general circulation at high altitude thathas high- and low-frequency content. A convection episode lastsaround a month, so variability should be created on the monthlytime scale—for example, by generating low-frequency planetarywaves that modulate the midlatitude jet stream (66, 67). Theinput into the upper troposphere is the cumulative result oftransient convection events that take a few days to rise and fall.Their space–time variability generates monthlong trains ofhigher-frequency waves with periods of 2.5 to 6 d (66). In such

Fig. 1. ENSO-related diagnostics, 1980 to 2017. Time-longitude Hovmuller diagrams of monthly anomalies (relative to P1 = 1980 to 1998 monthly clima-tology) of SSH (cm) (Upper Left), SLP (Pa) (Upper Right), SST (°C) (Lower Left), and the zonal component of the 10-m wind velocity (SSU, m/s) (Lower Right) areshown. SST, SSU, and SLP were derived by averaging the corresponding NCAR/NCEP Reanalysis data over the ±7.5° latitude belt and plotted as a function oflongitude in the equatorial Pacific Sector from 135° E to 90° W (vertical axis) with time extending from 1980 to 2017 (horizontal axis). SSH was derived byaveraging SSH relative to the geoid from the NCEP GODAS Reanalysis over the same latitude belt. The black line traces the time history of the longitude of the29 °C isotherm.

†A static instability region develops south of the ice edge in the Atlantic-facing sector(90°E to 0° to 90°W) in September and persists until at least December in P1 and P2 data.This unstable area is probably not due to ocean warming following reduction in Arcticsea-ice area since the calculations show instability in regions south of the ice edge inNovember and December where sea ice is found only in iceberg form.

4 of 11 | www.pnas.org/cgi/doi/10.1073/pnas.1717707117 Kennel and Yulaeva

Dow

nloa

ded

by g

uest

on

Apr

il 16

, 202

0

cases, the planetary waves would provide most of the momentumtransport (68), but the higher-frequency wave trains act to re-duce the Equator-to-pole temperature gradient. Wave transportcontinues in the upper troposphere, even after sea-ice regrowthfrustrates convection over the Arctic Ocean. Midlatitude weather isaffected 1 to 2 mo later (33, 69–71), though exactly how has beencontroversial (72).

P2 Arctic ITCZ-Teleconnections, Difference between P1 and P2Teleconnections, and Southward Advance of P2 Teleconnectionsare concerned with the planetary wave/wave-train interactionwith the ITCZ air column. Except in the jet streams and ITCZair column, the circulation in the upper troposphere is quasi-geostrophic (73); that is, dissipation- and divergence-free onconstant pressure surfaces; the wind velocity is derivable from ascalar function—the stream function—on those surfaces. Streamfunction contours characterize the teleconnections established bythe combined actions of the low-frequency planetary wave andthe higher-frequency wave train.Figs. 4–6 present the results of regressions of the stream

function on the SSIA. All three figures derive from NCAR/NCEP reanalysis data and have the same format. The 300-mbpressure surface was chosen as representative of upper tropo-sphere conditions; calculations not shown exhibit qualitativelyconsistent behavior for the 100- to 500-mb pressure surfaces.Monthly regression results for 300 mb are presented as anoma-lies relative to P1 climatology for both P1 and P2 data; the P1 SDin SSIA is the measure of sea-ice variability for both P1 and P2.All three figures show the response to a decrease in P2 SSIA ofone 1 P1 SD.The P2 December stream function response map (Fig. 4,

Upper) has three closed-contour cells that extend across thePacific to the tropics. The boxes locate the areas N4 and N3 usedto calculate the trade-wind dipole index (P1 and P2 El NinoEvents). The northernmost (and strongest) cell originates overthe East Siberian Sea–Barents Sea coastal arc that contains theregion of observed maximum sea-ice retreat and delay of fallrefreeze (74). Fig. 3 indicates that convective instability startsand ends in this region. This cell approaches the northernboundary of N3, which is near the typical latitude of the ITCZ in

P1: Aug 26 – Sep 05 P1: Sep 20 – Sep 30 P1: Oct 05 – Oct 15

P2: Sep 05 – Sep 15 P2: Oct 05 – Oct 15 P2: Oct 20 – Nov 04

Fig. 3. Time evolution of convective instability regions. ERA-interim datawere used to map atmospheric static stability at 80-m altitude above theArctic Ocean for P1 and P2. For each period, a series of 23 11-d averages with5-d moving centers running from August 1 to December 1 was created.Upper displays P1 maps for August (Aug) 26 to September (Sep) 5 (Left),September 20 to 30 (Center), and October (Oct) 5 to 15 (Right). Lower dis-plays P2 maps for September 5 to 15 (Left), October 5 to 15 (Center), andOctober 20 to November (Nov) 4 (Right). The black lines in each map denotethe P1 or P2 average September sea-ice extent. The polar plot ranges are 90°to 67° North latitude, and the units are degrees Kelvin per kilometer. Bluedenotes instability.

Fig. 2. Lagged correlation between the September sea-ice concentrationand the December trade-wind dipole index. Maps show the correlationbetween the December trade-wind dipole index and sea-ice concentra-tion (SIC) during the preceding September for P1 (Upper) and P2 (Lower).SIC data were taken from the NOAA/NSIDC passive microwave datarecord. The southernmost latitude is 67° North. Thick line denotes 15%SIC contour.

Kennel and Yulaeva PNAS Latest Articles | 5 of 11

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Apr

il 16

, 202

0

the EP, about 7.5°N (75). The second, more southerly cell ap-proaches the northern boundary of N4 in the CP, again, close tothe expected latitude of the ITCZ. The third cell connects SouthAsia to the southern boundary of N4 in the Western Pacific.The three cells have alternating signs: Their stream functions

are positive–negative–positive north to south. Since flow circu-lates counterclockwise (cyclonically) around a stream functionmaximum and clockwise (anticyclonically) around a minimum inthe Northern Hemisphere, conservation of angular momentumimplies that anomalies with these circulation senses would becommunicated to the surface winds below. The regressed 10-mwind anomalies in N3 and N4 are consistent with the signs of the300-mb stream-function anomalies at the northern boundaries ofN3 and N4. Fig. 4, Lower shows 10-m wind vectors superposedon color contours of the zonal wind-velocity anomalies calcu-lated for the same regression parameters as in Fig. 4, Upper. Thewind-anomaly vectors are easterly (blue) in N4 and mostlywesterly (red) in N3, indicating that the trade-wind dipoleanomaly is created when the anomalous upper troposphericcirculation in the two northernmost cells interacts with the ITCZair column.

Difference between P1 and P2 TeleconnectionsThe P1 and P2 atmospheres had different structures (76) andtherefore different teleconnection pathways. Fig. 5 compares theP1 (Left) and P2 (Right) regressions on SSIA of the December300-mb stream function (Top), 10-m zonal velocity (Middle), and

SST (Bottom). The P1 analog of the P2 northern teleconnectioncell connects to the region of maximum sea-ice loss as in P2, butturns eastward at midlatitudes and does not reach the ITCZ.This teleconnection pattern signifies a remote sea-ice influenceon midlatitude North American circulation, but not on tropicalPacific circulation. The P1 stream function has the same signover the northern boundaries of N3 and N4, implying thereshould be no trade-wind dipole anomaly. Accordingly, theregressed 10-m zonal wind velocities in Fig. 5, Left Middle havelargely the same sign in N3 and N4.The P1 and P2 regression responses of December SST are

shown in Fig. 5, Bottom. In P1 data, an SSIA decrease generatedsmall SST responses everywhere except in the Atlantic GulfStream near the North American East Coast. In P2 data, anSSIA decrease produced cooling responses in N4 and N3 andwarming responses in both the WPWP and the midlatitudeNorth Pacific. Note that an SSIA decrease of the same magni-tude (−1 P1 SD) drove both the P1 and P2 regressions, yet therewere significant trade-wind and temperature anomalies only inP2 data. This is an indication that the teleconnections betweenthe Arctic and the Tropics were stronger in P2 than in P1.

Southward Advance of P2 TeleconnectionsAccording to Initiation, Duration, and Spatial Extent of Convec-tion Episodes and P2 Arctic-ITCZ Teleconnections, convectionepisodes generate planetary waves with periods of about a monthand wave trains of about a month’s duration that take severalmonths to propagate from the Arctic to the Tropical Pacific. Ifso, the September-to-December sequence of monthly stream-function regression responses should provide low-resolutionsnapshots of the combined impacts of planetary wave modifica-tion of the jet streams and the southward advance of thewave trains.The P2 300-mb stream-function regression responses mapped

in Fig. 6 show an orderly series of changes from September toDecember. In P2 September, the stream-function cells over theNorthern Pacific are zonally aligned; there is no streamlineconnectivity between the Arctic and lower latitudes. The P2October stream-function anomaly is not organized intosouthward-extending structures; except for a small region east ofJapan, no streamline connects the Arctic with the Pacific. In P2November, the three cells identified in Fig. 4 are organized, andthere is streamline connectivity between the Arctic and theNorthern Pacific. The northern positive cell extends along theNorth American coast and turns eastward at midlatitudes, whilethe negative cell lies to its south. The negative cell does notapproach the ITCZ, and the stream functions over N3 and N4have the same sign, implying no trade-wind dipole anomaly. ByP2 December, the P2 November northern cell arrives at theITCZ in N3, its southern neighbor approaches it over N4, andthe trade-wind dipole anomaly appears in 10-m wind data.In summary, P2 September has no 300-mb stream-function

structures that connect the Arctic to the Tropical Pacific; P2December has structures that do link the Arctic to the ITCZ; andP2 October and P2 November are intermediate cases—a se-quence consistent with wave-train advance from the Arctic to theTropical Pacific.

Westerly Wind Bursts and the CP El NinoAccording to Fig. 1, eastward displacement of the 29 °C isothermto the CP sets the stage for both EP and CP El Nino events.Once the stage is set, westerly wind bursts (WWBs) (77–80)trigger them. WWBs are transient wind anomalies of 5- to 20-d duration that temporarily reverse or reduce the velocity of theeasterly trade winds; while not every WWB triggers an El Ninoevent, WWBs were associated with every EP and CP El Ninoevent between 1971 and 2010 (81).

P2: Dec 300 mb Stream Function

P2: Dec 10m zonal velocity and wind vectors

5 m/s

Fig. 4. Three-cell Arctic-Pacific telecommunication pattern. Upper shows aregression map of the 300-mb stream function anomalies (m2/s × 10−6) in P2December on the time series of the preceding P2 September Arctic sea-icearea anomalies. All anomalies are with respect to P1 climatology, and theregressions are scaled to a (−1) SD reduction of the P1 Arctic SSIA. Lowersuperposes regressed atmospheric 10-m wind-vector anomalies (blue east-erly and red westerly) on the regressed 10-m zonal velocity anomaly (m/s)(color contours). The boxes locate the regions N3 and N4 used to define thetrade-wind dipole anomaly. The arrow lengths at the bottom denote windvelocity of 5 m/s, with easterly in blue and westerly in red. The solid contoursenclose areas where regression coefficients differ from zero with more than90% confidence according to Student’s t test.

6 of 11 | www.pnas.org/cgi/doi/10.1073/pnas.1717707117 Kennel and Yulaeva

Dow

nloa

ded

by g

uest

on

Apr

il 16

, 202

0

Fig. 1 indicates that CP El Ninos are triggered near a point ofmaximum eastward displacement of the 29 °C isotherm that ef-fectively locates the eastern boundary of the WPWP. The tran-sition layer between Warm Pool and CP water is characterized bystrong zonal gradients in SST, SSH, and SSS; the CP El Nino isassociated with relaxation of those gradients (82). By reducingtrade-wind stress on the ocean surface, a WWB over the tran-sition layer releases a shallow eddy-mixed layer of higher-SSH,higher-SST, lower-salinity, lower-density water eastward undergravity; after triggering, WPWP water advects over the lower-

SSH, lower-SST, higher-SSS, higher-density CP water (83) tothe east of the transition layer. There is little delay between theWWB trigger and the appearance of the positive SST anomalythat moves eastward from the point of triggering.WWB triggering occurs especially in the months that wave

trains generated by Arctic convection episodes interact with theITCZ air column, since maximum CP El Nino occurrence is inDecember and January (23). This temporal coincidence moti-vates two questions. First, are WWBs produced during that in-teraction? Second, since a larger SSH gradient makes it more

P1: Dec 300 mb Stream Function P2: Dec 300 mb Stream Function

P1: Dec 10m Zonal Winds P2: Dec 10m Zonal Winds

P1: Dec SST P2: Dec SST

Fig. 5. Regression coefficients of P1 and P2 variables against SSIA. Top compares the December (Dec) 300-mb stream-function structures generated byregression on seasonal sea-ice area in P1 (Left) and P2 (Right). Middle shows the regressed zonal component of the 10-m wind velocity, in units of m/s, for P1(Left) and P2 (Right). Bottom shows the regressed SST (°C) for P1 (Left) and P2 (Right). The data source, formatting, regression parameters, and contour linesare the same as for Fig. 4.

P2: Sep P2: Oct

P2: Nov P2: Dec

Fig. 6. Monthly evolution of upper troposphere teleconnections. Shown are regression maps (m2/s × 10−6) for September (Sep; Upper Left), October (Oct;Upper Right), November (Nov; Lower Left), and December (Dec; Lower Right) of the 300-mb stream function on the time series of the preceding SeptemberArctic sea-ice area. Regression conditions, formats, and contour-line meaning are the same as in Fig. 4, Upper.

Kennel and Yulaeva PNAS Latest Articles | 7 of 11

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Apr

il 16

, 202

0

likely that a WWB of a given amplitude will trigger a CP El Ninoevent, does the wave-train interaction with the ITCZ air columnincrease the SSH gradient in the transition layer?The 1-mo time resolution of SSIA regression methods cannot

diagnose the wave-train interaction with the ITCZ air column,but it is possible to infer certain properties of the interactionusing reasoning based upon earlier results in this paper. The areaof the Pacific-facing sector of the Arctic Ocean that is unstable toatmospheric convection increases from zero, maximizes, andreturns to zero over a 40- to 45-d period (Initiation, Duration,and Spatial Extent of Convection Episodes). An Arctic convectionepisode should therefore comprise a roughly 20-d period ofstrengthening convection, followed by a 20-d period of weaken-ing convection. The wave trains traveling southward from theArctic should also strengthen and weaken over comparable pe-riods. During wave-train interaction with the ITCZ air column,the easterly wind speed in the CP should increase, maximize, anddecrease (Correlation of September Sea Ice Concentration withDecember Trade-Wind Dipole Anomaly Index), and the decreasingphase could be interpreted as a WWB.Regression of SSH on SSIA provides indirect evidence about

the modification of the SSH gradient by the wave-train in-teraction with the ITCZ air column. Fig. 7 maps color contoursof SST (Left) and SSH (Right) regressed to −1 P1 SD reductionin SSIA for P2 November (Top), P2 January (Middle), and P2February (Bottom). The boxes locate the areas N3 and N4 thatdefine the trade-wind dipole anomaly index (P1 and P2 El NinoEvents). Samples of regressed 10-m wind vectors are mapped onthe SST and SSH color contours. The SSH gradient may be vi-sually estimated from the spacing of the SSH contours.In the model in Conceptual Model, Methods, and Data, a wave

train of about a month’s duration approaches the ITCZ in No-vember (Fig. 6), arrives at the ITCZ in December (Figs. 4 and 6),completes its interaction with the ITCZ air column in January

(Fig. 7), and launches a reflected wave train that arrives over theAleutian Low Circulation in February (Fig. 8). Since P1 wavetrains do not reach the ITCZ (Fig. 5), the P1 SSH, SST, and windregressions show little change between P1 November, P1 Janu-ary, and P2 February and are not shown. The SSH results for P2do show significant change. For P2 November, Fig. 7 shows amild SSH gradient at the western boundary of N4, where CP ElNinos are triggered (Fig. 1). A positive SSH anomaly appears inthe Eastern Indian Ocean and in the WPWP in P2 January; theSSH gradient near the western boundary of N4 is strongest inthis month. In P2 February, the SSH anomaly in the EasternIndian and Western Pacific Oceans is smaller than in P2 January,and the SSH gradient near the western boundary of N4 is smallerthan in P2 January.Fig. 7 also contains SSH evidence that processes within the

ITCZ respond to sea-ice variability in P2. There is a thin band ofnegative SSH regression response near the expected position ofthe ITCZ at the northern edge of N3 in P2 November, P2 Jan-uary, and P2 February. Regression also detects a parallel struc-ture south of the Equator in N3 in P2 January and P2 February,months that precipitation observations find that a double ITCZoccurs (84). This pair of negative SSH anomalies encloses anegative SST anomaly localized to the geographic equator inthese months. Further evidence would be required to confirmthat a double ITCZ is a delayed response to reduction in SSIA.In summary, in P2 data, reduction in SSIA regresses to SSH

zonal gradients in the WPWP transition layer favorable to ad-vective instability in the months when convection episode-created wave trains interact with the ITCZ air column. A re-duction in trade-wind speed in the diminishing phase of thewave-train interaction would appear as a WWB lasting around20 d. Fig. 1 indicates that that summer Arctic sea-ice loss doesnot couple to CP events in years when the 29 °C isotherm is inthe far Western Pacific. We do not argue that all WWBs and/or

P2: Nov SST and 10m winds P2: Nov SSH and 10m winds

P2: Jan SST and 10m winds P2: Jan SSH and 10m winds

P2: Feb SST and 10m winds P2: Feb SSH and 10m winds

5 m/s5 m/s

Fig. 7. North Pacific P2 SST, SSH, and 10-m wind regressions. Shown are November (Nov) SST (Top Left), November SSH (Top Right), January (Jan) SST (MiddleLeft), January SSH (Middle Right), February SST (Bottom Left), and February (Feb) SSH (Bottom Right). The SSH color scale ranges from −5 to 5 dm; the SSTcolor ranges from −5 to 5 °C. The arrow lengths at the bottom of each panel denote wind velocity of 5 m/s, with easterly in blue and westerly in red. The 10-mwinds and SST were drawn from NCEP/NCAR and SSH from GODAS data. Regression parameters, N3 and N4 boxes, and contour-line meaning are the same asin Fig. 4.

8 of 11 | www.pnas.org/cgi/doi/10.1073/pnas.1717707117 Kennel and Yulaeva

Dow

nloa

ded

by g

uest

on

Apr

il 16

, 202

0

CP El Nino events require an Arctic-origin trigger. We do arguethat summer sea-ice loss led to SSH gradients more favorable toadvection instability in P2 and less favorable in P1 (when sum-mer sea losses were smaller and CP events were not observed asfrequently); that SSH gradients favorable to advective instabilityare generated under regression near where CP events originate;that the switch each fall from sea-ice retreat to advance providesfor a westerly wind anomaly on the 20-d time scale in winter; andthat those CP events that are Arctic sea-ice-related should beinitiated in December and January.

Arctic Sea Ice and the Aleutian Low CirculationThe Aleutian Low Circulation is the dominating feature of winterNorth Pacific climate, organizing storm systems and the jet stream(85). The Aleutian Low Circulation changed character in the late1990s when CP El Ninos began to occur more frequently (86).This section examines the hypothesis that waves stimulated byconvection episodes generate reflected waves in interaction withthe ITCZ air column that modify the Aleutian Low Circulation.Waves reflected from the ITCZ in December should reach the

latitude of the Aleutian Low Circulation in February. Fig. 8,Upper shows the regressed 300-mb stream function for P1 Feb-ruary and P2 February. The P1 stream-function map has quasi-periodic zonal structure, whereas the P2 stream-function maphas three Pacific-wide zonally aligned cells. The P2 Februaryparity sequence is the reverse of the P2 December sequence (Fig.4), negative–positive–negative north to south, consistent with anArctic-origin wave train reflected in interaction with the ITCZ aircolumn. The zonal extent of the P2 teleconnection cells suggeststhat reflected waves are generated along much of the ITCZ.Fig. 8, Lower shows regressed 10-m wind vectors mapped over

contours of SLP for P1 February (Left) and P2 February (Right).In the tropics, both SLP and surface winds are unaffected bySSIA variability in P1, whereas the P2 regression response is anincrease (decrease) of SLP in the tropical Eastern (Western)Pacific. The difference in the Northern Pacific between P1February and P2 February is notable. Smaller SSIA in P1 leadsto lower SLP across the midlatitudes and a dipolar SLP anomalyfurther north. In P2 February, smaller SSIA leads to a singlepositive SLP anomaly. Its accompanying anticyclonic surfacewind anomaly weakens the prevailing cyclonic Aleutian LowCirculation.

A significant response of the Aleutian Low Circulation toSSIA variability occurs only in P2 February data. The regressionresponses for P2 January and P2 March (not presented here) aremuch weaker. The off–on–off January–February–March se-quence is consistent with an Arctic-origin wave train of about amonth’s duration that reflects from the ITCZ air column inDecember, propagates poleward in January, passes over theAleutian Low Circulation in February, and enters the Arcticin March.

RemarksTwo limitations of technique prevent this paper’s results frombeing conclusive. The first limitation stems from partitioningdata into smaller subsets, risking the loss of statistical precision.For example, not separating P1 from P2 data would obscurethe effects of the different teleconnection pathways linkingArctic sea ice and Pacific trade-wind variability in the two pe-riods. The further partitioning of P1 and P2 data into monthlysubsets again sacrifices precision in favor of temporal discrimi-nation. If one accepts these risks as necessary for the problem athand, one encounters a second limitation, circularity: To test themodel, one has to assume the model. The credibility of the analysisof an individual step in the model in Conceptual Model, Methods,and Data depends on the credibility of the analyses of other stepsin the model. The best that can be said is that at the level ofprecision attainable with the present methods, there is no obvi-ous inconsistency between the results and the model used to in-terpret the results.Regression has identified climatic processes that relate di-

rectly or indirectly to the variability of September Arctic sea-ice area. These include the trade-wind dipole anomaly, theAleutian Low Circulation, the CP El Nino, and (possibly) thedouble ITCZ. Regression also identifies upper- tropospheretelecommunication pathways connecting Arctic and tropicsthat could account for the regression responses at the surface.Proof that CP El Nino events triggered by Arctic origin wavetrains can prompt secondary teleconnections could encouragea new view of the teleconnection network that communicatesshort-term climatic changes around the globe. When the ElNino feedback system is close to marginal stability, a smallincoming telecommunication signal could trigger an SST re-sponse that modulates the vertical convection in the ITCZ

s/m5s/m5

Fig. 8. Northern Pacific teleconnections for the month of February (Feb). Upper contours the regressed 300-mb stream function in P1 (Left) and P2 (Right).The regression conditions, boxes, and contour-line meanings are the same as in Figs. 4–6. Lower shows selected 10-m wind vectors superposed on SLP. NCEP/NCAR data were used. The arrow lengths at the bottom denote wind velocity of 5 m/s, with easterly in blue and westerly in red.

Kennel and Yulaeva PNAS Latest Articles | 9 of 11

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Apr

il 16

, 202

0

air column. The deep convection generated in the ITCZ aircolumn is the main power source of the Hadley Cell over-turning circulations and indirectly of global atmospheric cir-culation; by amplifying (or damping) teleconnection signals,the ITCZ air column becomes an active participant in theglobal teleconnection network.Whether teleconnection signals are amplified or damped during

their interactions with the ITCZ air column depends on thestate of the Tropical Pacific Ocean, which depends on its history.Calculating rates of amplification or damping requires calculatingthe response of the ITCZ air column to wave-train forcing in theupper troposphere and stratosphere while exchanging energy withthe subsurface ocean. Every winter provides an opportunity to

make observational progress on this extraordinarily complexproblem. It could prove worth the effort. ITCZ amplification ofteleconnection signals could help explain how the loss of NorthernHemisphere land ice resulted in massive worldwide climate changeafter the Last Glacial Maximum (87).

ACKNOWLEDGMENTS. V. Ramanathan, K.K. Tung, and S.-P. Xie reviewedearly drafts of this paper; they are not responsible for the errors that remain.We thank M. B. Kennel, M. J. Rees, E. Shuckburgh, and I. Zaslavski for helpfulinsights. We thank the reviewer who carefully pointed out where there wasneed for clearer presentation and our PNAS editor. C.F.K. was supported byChrist’s College, Cambridge, and the University of California. E.Y. was sup-ported by the Scripps Institution of Oceanography and the San DiegoSupercomputer Center.

1. S. Manabe, R. J. Stouffer, Sensitivity of a global climate model to an increase of CO2

concentration in the atmosphere. J. Geophys. Res. Oceans 85, 5529–5554 (1980).2. J. A. Screen, I. Simmonds, The central role of diminishing sea ice in recent Arctic

temperature amplification. Nature 464, 1334–1337 (2010).3. J. A. Francis, S. J. Vavrus, Evidence linking Arctic amplification to extreme weather in

mid-latitudes. Geophys. Res. Lett. 39, 6 (2012)4. J. Cohen et al., Recent Arctic amplification and extreme mid-latitude weather. Nat.

Geosci. 7, 627–637 (2014).5. Arctic Climate Impact Assessment, Impacts of a Warming Arctic-Arctic Climate Impact

Assessment, by Arctic Climate Impact Assessment (Cambridge University Press, Cam-

bridge, UK, 2004).6. Arctic Monitoring and Assessment Program, Snow, Water, Ice and Permafrost in the

Arctic (SWIPA): Climate Change and the Cryosphere (Arctic Monitoring and Assess-

ment Program, Oslo, Norway, 2011).7. K. Pistone, I. Eisenman, V. Ramanathan, Observational determination of albedo de-

crease caused by vanishing Arctic sea ice. Proc. Natl. Acad. Sci. U.S.A. 111, 3322–3326

(2014).8. R. Zhang, H. Wang, Q. Fu, P. J. Rasch, X. Wang, Unraveling driving forces explaining

significant reduction in satellite-inferred Arctic surface albedo since the 1980s. Proc.

Natl. Acad. Sci. U.S.A. 116, 23947–23953 (2019).9. F. Pithan, T. Mauritsen, Arctic amplification dominated by temperature feedbacks in

contemporary climate models. Nat. Geosci. 7, 181–184 (2014).10. M. O. Jeffries, J. E. Overland, D. K. Perovich, The Arctic shifts to a new normal. Phys.

Today 66, 35–40 (2013).11. K. E. Trenberth, J. T. Fasullo, An apparent hiatus in global warming? Earths Futur. 1,

19–32 (2013).12. C. F. Kennel, “Hiatus” in global warming: Paradox and complexity in climate science.

Proc. Am. Philos. Soc. 159, 367 (2015).13. D. G. Victor, C. F. Kennel, Climate policy: Ditch the 2 °C warming goal. Nature 514, 30–

31 (2014).14. M. A. Balmaseda, K. E. Trenberth, E. Källén, Distinctive climate signals in reanalysis of

global ocean heat content. Geophys. Res. Lett. 40, 1754–1759 (2013).15. G. A. Meehl, A. Hu, J. M. Arblaster, J. Fasullo, K. E. Trenberth, Externally forced and

internally generated decadal climate variability associated with the Interdecadal

Pacific Oscillation. J. Clim. 26, 7298–7310 (2013).16. M. H. England et al., Recent intensification of wind-driven circulation in the Pacific

and the ongoing warming hiatus. Nat. Clim. Chang. 4, 222–227 (2014).17. Y. Kosaka, S.-P. Xie, Recent global-warming hiatus tied to equatorial Pacific surface

cooling. Nature 501, 403–407 (2013).18. M. Watanabe et al., Contribution of natural decadal variability to global warming

acceleration and hiatus. Nat. Clim. Change 4, 893–897 (2014).19. X. Chen, K.-K. Tung, Climate. Varying planetary heat sink led to global-warming

slowdown and acceleration. Science 345, 897–903 (2014).20. K. Ashok, T. Yamagata, Climate change: The El Niño with a difference. Nature 461,

481–484 (2009).21. S.-R. Yeo, K.-Y. Kim, Decadal changes in the southern hemisphere sea surface tem-

perature in association with El Nino-Southern Oscillation and Southern Annular

Mode. Clim. Dyn. 45, 3227–3242 (2015).22. T. Lee, M. J. McPhaden, Increasing intensity of El Niño in the central-equatorial Pacific.

Geophys. Res. Lett. 37, L14603 (2010).23. H.-Y. Kao, J.-Y. Yu, Contrasting Eastern-Pacific and Central-Pacific types of ENSO. J.

Clim. 22, 615–632 (2009).24. C. Fu, H. Diaz, J. Fletcher, Characteristics of the response of sea surface temperature in

the central Pacific associated with warm episodes of the Southern Oscillation. Mon.

Weather Rev. 114, 1716–1739 (1986).25. K. E. Trenberth, D. P. Stepaniak, Indices of El Niño evolution. J. Clim. 14, 1697–1701

(2001).26. S. W. Yeh et al., El Niño in a changing climate. Nature 461, 511–514 (2009).27. M. B. Freund et al., Higher frequency of Central Pacific El Niño events in recent de-

cades relative to past centuries. Nat. Geosci. 12, 450–455 (2019).28. M. Feng et al., Decadal increase in Ningaloo Niño since the late 1990s. Geophys. Res.

Lett. 42, 104–112 (2015).29. D. Coumou, S. Rahmstorf, A decade of weather extremes. Nat. Clim. Change 2, 491–

496 (2012).

30. Intergovernmental Panel on Climate Change, Managing the Risks of Extreme Eventsand Disasters to Advance Climate Change Adaptation, C. B. Field et al., Eds. (Cam-bridge University Press, Cambridge, UK, 2012), pp. 1–19.

31. J. Liu, J. A. Curry, H. Wang, M. Song, R. M. Horton, Impact of declining Arctic sea iceon winter snowfall. Proc. Natl. Acad. Sci. U.S.A. 109, 4074–4079 (2012).

32. C. Hong et al., Roles of European blocking and tropical–extratropical interaction inthe 2010 Pakistan flooding. Geophys. Res. Lett. 38, L13806 (2011).

33. S. J. Vavrus, The influence of Arctic amplification on mid-latitude weather and cli-mate. Curr. Clim. Change Rep. 4, 238–249 (2018).

34. J. A. Screen et al., Consistency and discrepancy in the atmospheric response to Arcticsea-ice loss across climate models. Nat. Geosci. 11, 155–163 (2018).

35. R. Blackport, J. A. Screen, K. van der Wiel, R. Bintanja, Minimal influence of reducedArctic sea ice on coincident cold winters in mid-latitudes. Nat. Clim. Chang. 9, 697–704(2019).

36. D. Budikova, T. W. Ford, T. J. Ballinger, United States heat wave frequency and ArcticOcean marginal sea ice variability. J. Geophys. Res. D Atmos. 124, 6247–6264 (2019).

37. P. Zhang et al., A stratospheric pathway linking a colder Siberia to Barents-Kara Seasea ice loss. Sci. Adv. 4, eaat6025 (2018).

38. J. C. Fyfe et al., Making sense of the early-2000s warming slowdown. Nat. Clim.Chang. 6, 224–228 (2016).

39. J. C. Stroeve et al., Trends in Arctic sea ice extent from CMIP5, CMIP3 and observa-tions. Geophys. Res. Lett. 39, L06801 (2012).

40. T. Koenigk, A. Devasthale, K. G. Karlsson, Arctic sea ice albedo in CMIP5 models.Atmos. Chem. Phys. 14, 1987–1998 (2014).

41. D. Olonscheck, T. Mauritsen, D. Notz, Arctic sea-ice variability is primarily driven byatmospheric temperature fluctuations. Nat. Geosci. 12, 430–434 (2019).

42. M. L. Kapsch, N. Skific, R. G. Graversen, M. Tjernström, J. A. Francis, Summers with lowArctic sea ice linked to persistence of spring atmospheric circulation patterns. Clim.Dyn. 52, 2497–2512 (2019).

43. I. Cvijanovic et al., Future loss of Arctic sea-ice cover could drive a substantial decreasein California’s rainfall. Nat. Commun. 8, 1947 (2017).

44. S. Matsumura, Y. Kosaka, Arctic-Eurasian climate linkage induced by tropical oceanvariability. Nat. Commun. 10, 3441 (2019).

45. E. Kalnay et al., The NCEP/NCAR 40-year reanalysis project. Bull. Am. Meteorol. Soc.77, 437–471 (1996).

46. R. Kistler et al., The NCEP–NCAR 50-year reanalysis: Monthly means CD-ROM anddocumentation. Bull. Am. Meteorol. Soc. 82, 247–267 (2001).

47. D. P. Dee et al., The ERA‐Interim reanalysis: Configuration and performance of thedata assimilation system. Q. J. R. Meteorol. Soc. 137, 553–597 (2011).

48. E. Jakobson et al., Validation of atmospheric reanalyses over the central Arctic Ocean.Geophys. Res. Lett. 39, L10802 (2012).

49. R. Lindsay et al., Evaluation of seven different atmospheric reanalysis products in theArctic. J. Clim. 27, 2588–2606 (2014).

50. W. N. Meier et al., NOAA/NSIDC Climate Data Record of Passive Microwave Sea IceConcentration, Version 3 (NSIDC, Boulder, CO, 2017).

51. G. Peng, W. N. Meier, D. Scott, M. Savoie, A long-term and reproducible passive mi-crowave sea ice concentration data record for climate studies and monitoring. EarthSyst. Sci. Data 5, 311–318 (2013).

52. D. W. Behringer, Y. Xue, “Evaluation of the global ocean data assimilation system atNCEP: The Pacific Ocean” in Proceedings. Eighth Symposium on Integrated Observingand Assimilation Systems for Atmosphere, Oceans, and Land Surface, AMS 84thAnnual Meeting . https://www.esrl.noaa.gov/psd/cgi-bin/db_search/DBListFiles.pl?did=98&tid=72048&vid=1916. Accessed 15 March 2019.

53. E. Hovmöller, The trough-and-ridge diagram. Tellus 1, 62–66 (1949).54. N. E. Graham, T. P. Barnett, Sea surface temperature, surface wind divergence, and

convection over tropical oceans. Science 238, 657–659 (1987).55. N. C. Johnson, S. P. Xie, Changes in the sea surface temperature threshold for tropical

convection. Nat. Geosci. 3, 842–845 (2010).56. T. Qu, Y. T. Song, C. Maes, Sea surface salinity and barrier layer variability in the

equatorial Pacific as seen from Aquarius and Argo. J. Geophys. Res. Oceans 119, 15–29(2014).

57. T. Qu, J.-Y. Yu, ENSO indices from sea surface salinity observed by Aquarius and Argo.J. Oceanogr. 70, 367–375 (2014).

58. K. E. Trenberth, The definition of El Nino. Bull. Am. Meteorol. Soc. 78, 2771–2777(1997).

10 of 11 | www.pnas.org/cgi/doi/10.1073/pnas.1717707117 Kennel and Yulaeva

Dow

nloa

ded

by g

uest

on

Apr

il 16

, 202

0

59. J. Bjerknes, Atmospheric teleconnections from the equatorial Pacific. Mon. WeatherRev. 97, 163–172 (1969).

60. S. E. Zebiak, M. A. Cane, A model El Niño–Southern Oscillation. Mon. Weather Rev.115, 2262–2278 (1987).

61. A. V. Fedorov et al., The impact of westerly wind bursts and ocean initial state on thedevelopment, and diversity of El Niño events. Clim. Dyn. 44, 1381–1401 (2015).

62. H. Bi et al., Contributions of advection and melting processes to the decline in sea icein the Pacific sector of the Arctic Ocean. Cryosphere 13, 1423–1439 (2019).

63. R. Jaiser et al., Impact of sea ice cover changes on the Northern Hemisphere atmo-spheric winter circulation. Tellus A Dyn. Meteorol. Oceanogr. 64, 11595 (2012).

64. J. E. Kay, A. Gettelman, Cloud influence on and response to seasonal Arctic sea iceloss. J. Geophys. Res. D Atmos. 114, D18204 (2009).

65. H. Wernli, L. Papritz, Role of polar anticyclones and mid-latitude cyclones for Arcticsummertime sea-ice melting. Nat. Geosci. 11, 108 (2018).

66. J. A. Francis, S. J. Vavrus, Evidence for a wavier jet stream in response to rapid Arcticwarming. Environ. Res. Lett. 10, 014005 (2015).

67. D. Coumou, V. Petoukhov, S. Rahmstorf, S. Petri, H. J. Schellnhuber, Quasi-resonantcirculation regimes and hemispheric synchronization of extreme weather in borealsummer. Proc. Natl. Acad. Sci. U.S.A. 111, 12331–12336 (2014).

68. V. Limpasuvan, D. L. Hartmann, Wave-maintained annular modes of climate vari-ability. J. Clim. 13, 4414–4429 (2000).

69. C. Deser, R. A. Tomas, S. Peng, The transient atmospheric circulation response to northAtlantic SST and sea ice anomalies. J. Clim. 20, 4751–4767 (2007).

70. J. A. Francis, W. Chan, D. J. Leathers, J. R. Miller, D. E. Veron, Winter NorthernHemisphere weather patterns remember summer Arctic sea-ice extent. Geophys. Res.Lett. 36, L07503 (2009).

71. J. A. Francis, N. Skific, S. J. Vavrus, North American weather regimes are becomingmore persistent: Is Arctic amplification a factor? Geophys. Res. Lett. 45, 11–414 (2018).

72. J. M. Wallace, I. M. Held, D. W. Thompson, K. E. Trenberth, J. E. Walsh, Globalwarming and winter weather. Science 343, 729–730 (2014).

73. B. J. Hoskins, The geostrophic momentum approximation and the semi-geostrophicequations. J. Atmos. Sci. 32, 233–242 (1975).

74. J. C. Stroeve et al., Changes in Arctic melt season and implications for sea ice loss.Geophys. Res. Lett. 41, 1216–1225 (2014).

75. K. R. Wodzicki, A. D. Rapp, Long‐term characterization of the Pacific ITCZ usingTRMM, GPCP, and ERA‐Interim. J. Geophys. Res. D Atmos. 121, 3153–3170 (2016).

76. L. Yu, S. Zhong, M. Zhou, D. H. Lenschow, B. Sun, Revisiting the linkages between thevariability of atmospheric circulations and Arctic melt-season sea ice cover at multipletime scales. J. Clim. 32, 1461–1482 (2019).

77. D. S. Luther, D. E. Harrison, R. A. Knox, Zonal winds in the central equatorial Pacificand El Nino. Science 222, 327–330 (1983).

78. R. A. Kerr, Atmospheric science: Does a globe-girdling disturbance jigger El Nino?Science 285, 322–323 (1999).

79. M. J. McPhaden, Evolution of the 2002/03 El Nino. Bull. Am. Meteorol. Soc. 85, 677–695 (2004).

80. D. Chen et al., Strong influence of westerly wind bursts on El Niño diversity. Nat.Geosci. 8, 339 (2015).

81. T. Lian, D. Chen, Y. Tang, Q. Wu, Effects of westerly wind bursts on El Niño: A newperspective. Geophys. Res. Lett. 41, 3522–3527 (2014).

82. J. Qi et al., Salinity variability in the tropical Pacific during the Central-Pacific andEastern-Pacific El Niño events. J. Mar. Syst. 199, 103225 (2019).

83. J.-S. Kug, F.-F. Jin, S.-I. An, Two types of El Niño events: Cold tongue El Niño and warmpool El Niño. J. Clim. 22, 1499–1515 (2009).

84. S. P. Xie et al., Eastern Pacific ITCZ dipole and ENSO diversity. J. Clim. 31, 4449–4462(2018).

85. J. E. Overland, J. M. Adams, N. A. Bond, Decadal variability of the Aleutian low and itsrelation to high-latitude circulation. J. Clim. 12, 1542–1548 (1999).

86. S. W. Yeh, X. Wang, C. Wang, B. Dewitte, On the relationship between the NorthPacific climate variability and the central Pacific El Niño. J. Clim. 28, 663–677 (2015).

87. T. R. Jones et al., Southern Hemisphere climate variability forced by NorthernHemisphere ice-sheet topography. Nature 554, 351–355 (2018).

Kennel and Yulaeva PNAS Latest Articles | 11 of 11

EART

H,A

TMOSP

HER

IC,

ANDPL

ANET

ARY

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Apr

il 16

, 202

0